Add New Relic monitoring #16
Description
Percentile stats would be great (95th and/or 99th percentile for service times), as would visibility into individual request types. Right now we only have coarse POST vs. GET via log-deranged + librato, and only mean/max.
We should test first on a dev service with simulated high-rate POSTs. The agent warnings we've seen elsewhere make me nervous. We should look for warnings that the agent's connection has been cut off, and we should monitor the memory usage.